A Study on Estimation of Financial Liquidity Risk Prediction Model Using Financial Analysis

Transcription

1 A Study on Estimation of Financial Liquidity Risk Prediction Model Using Financial Analysis Chang-Ho An* *Department of Financial Information Engineering (Statistics), Seokyeong University, 124, Seokyeong-ro, Seongbuk-gu, Seoul 02713, Korea. * Correspond author: Chanh-Ho An, Ph.D. Abstract In this study we conducted a static financial analysis of the financial ratios of the manufacturing, information service, and financial and insurance industries to propose a model for predicting the financial risk, and performed ANOVA to select the significant variables affecting the healthy firms and the poor firms. Using the results of ANOVA, the linear discriminant model, the secondary discriminant model, the probit model, and the logit discriminant model were estimated and tested to propose a predictive model. The results of the test are as follows. The variables selected by ANOVA were all significant in the significance test of all models. And the fitness and the explanatory power of the estimated model were evaluated by the Apparent Error Rate (APER) which is the misclassification rate of the confusion matrix which is the classification matrix of the observation result and the prediction result. As the result, the misclassification rate of the logit discriminant model was the lowest, and the next lowest model was the probit model. Therefore, in order to compare the predictive power of the probit and logit discriminant models, the association analysis between the prediction probability and the observation response was analyzed and the predictive power was compared using rank correlation coefficients. As the result, the predictive power of the logit discriminant model is higher than that of the probit model. Therefore, the logit discrimination model, which has the lowest misclassification rate of all data and has the highest forecasting power, is proposed as the final model for predicting the financial risk. Keywords: ANOVA, Probit model, Logit discriminant model, Misclassification rate, Rank correlation coefficients INTRODUCTION Korean economy has cut its base rate several times in order to escape the economic crisis that started with the global financial crisis and has been maintaining its low interest rate policy. This is because there is a strong will and belief that the economy can be recovered only when money is released on the market. However, the rapid increase in market due to low interest rates should never be ignored. The sharp rise in can lead to a bubble problem by raising the market value too much, and the short-termed shortfall of market funds, which is released to the financial sector, may become serious. Such speculative short-term funds are likely to lead to another problem called the crisis, since there is a substantial amount of money but no substantial flow. Therefore, as the strategy to avoid the second economic crisis, some carefully express to raise interest rates. The international community believes that Korean economy is at risk of financial due to the continued low interest rates. In particular, the IMF points out that all Korean companies are facing corporate debt problems, and that all industries need to be restructured because of the increase in household debts and the problem of soundness of corporate debts, which are represented by insolvent companies. Recently the Federal Reserve Board (FRB) of the US raised its base interest rate as the first step for normalizing its monetary policy. If the US raises the interest rate, the increase in the risk is inevitable as the management condition deteriorates due to the rapid leaking of investment funds that has flowed into Korea and emerging countries. Therefore, it is necessary to continually study the forecasting models needed to manage the risk of companies, considering the sluggish sales of companies due to the slowdown in economic growth, low interest rate policy, interest rate hikes in the US, and uncertainties in overseas markets. Similar previous studies of the financial forecasting model are as follows. In 1985, West analyzed bank failures using factor analysis and logit analysis [1], and in 1991, Tam analyzed the banking industry using neural network analysis [2]. In 1994, Cho developed a logistic model predicting in the service industry [3], and in 1996, Lee et al. confirmed the predictive power of hybrid neural network models using 57 financial ratios [4]. In 1999, Davalos et al. predicted the risk of airlines by using neural network method, and in 1997, Altman and Narayanan suggested that the predictive power of each country varies according to the economic environment. In 2007, Davalos et al. suggested that activity indicators are important variables for risk prediction [7], and in 2009, Boyacioglu et al. used logit model and discriminant model to test the predictive 9919

2 power using 20 financial ratios [8]. In 2011, Kim emphasized that the growth indicators are important variables for risk prediction [9]. In 2015, Iturriaga and Sanz compared and analyzed the predictive power between models by performing discriminant analysis, logit analysis and neural network analysis on US commercial banks [10]. In 2016, An analyzed the ratios of unlisted companies and proposed a predicting model using multiple regression model [11]. As a previous research using logit and multiple regression models in other fields, in 2017, Kim used the multiple regression model to test the identity effect of body shape recognition and body shape management [12], and in 2017, Hwang and Kim analyzed the potential risk of infection using logistic analysis [13]. And in 2006, Tabachnick and Fidell found that in multivariate discriminant analysis the assumption of multivariate normality did not seem to be a big problem as far as it was not violated by extreme values, if the number of predictive variables was small and the number of cases in the smallest group was 20 or more, even though the numbers of cases in groups were different [14]. RESEARCH METHODS Data and research model The data used in this study are the financial statements of the manufacturing, information service, and financial and insurance industries in The total number of enterprises is 165; the healthy companies are 89, and the bad companies 76. The financial analysis variables used are Current ratio (CUR), Cash ratio (CR), Quick ratio (QR), Net working capital ratio (NR), Debt ratio (DR), borrowings and bonds payable to total assets (TA), Interest coverage ratio (IR), EBITDA to interest cost ratio (EIR), stockholders equity to total assets (SA), and Interest expenses to total borrowings and bonds payable (IP). As a research model, when f 1 (x) and f 2 (x) are probability density functions of healthy companies and bad companies respectively, the logit model considered is as follows: log f 1(x) f 2 (x) = β 0 + β x (2-1) The probabilities P 1 (x) and P 2 (x) belonging to two firms are estimated as follows: P 1 (x) = exp (β 0+β x) 1+exp (β 0 +β x) P 2 (x) = 1 P 1 (x) (2-2) Then, the prediction result is classified as follows using the classification reference value d = 0.5 β 0 + β x 0 d (2-3) β 0 + β x 0 < d Bad Assessment of the fit of the model The Apparent Error Rate (APER), which evaluates the model fit, is the ratio of misclassified data when the data is reclassified by using the estimated classification function. It is a measure of the misclassification rate of any classification function regardless of the distribution pattern of the group and is calculated as follows using the correct classification rate (CCR), APER = 1 CCR (2-4) where CCR = ( n ii+n i+ ) 100%, n is the total number of n observations, n ii is the diagonal element regularly stored in the classification matrix, and n i+ is the sum of ith row. Estimation of predictive power of the model The rank correlation coefficients, which is the statistic that evaluates the predictive power of probit and logit discriminant models, are calculated as follows. In the comparative model, a model with large values of rank correlation coefficients is judged as a model with good predictive power. Somer s D = 1 t (n c n d ) (2-5) Goodman Kruskal Gamma = n c n d n c +n d Kendall s Tau a = n c n d n(n 1)/2 C = 1 t {n c (t n c n d )} where n is the total number of observations, t is the total number of pairs having different response value, n c is the number of matched pairs, and n d is the number of mismatched pairs. RESULTS AND DISCUSSION The prior probability and classification reference value used in the comparative analysis model was 0.5, and it was confirmed that the misclassification rate in the linear discriminant model was 27.62%, higher than that in the secondary discriminant model. Therefore, a prediction model was proposed by comparatively analyzing the secondary discriminant model, the probit model, and the logit discriminant model. The comparison results are shown in Table 1 to Table 5. Variable selection and significance test Table 1 shows the results of ANOVA of identity test between two companies. Based on the F-test with the significance level of 5%, the variables CUR, CR, QR, NR, TA, and SA were selected as statistically significant variables for discriminating 9920

3 between healthy companies and bad companies. The remaining variables were excluded from the independent variables of the estimation model. Variables Table 1. ANOVA Bad Statistics Mean Std. dev. Mean Std. dev. F p- value CUR CR QR NR DR TA IR EIR SA IP Table 2 shows the results of testing the significance of the discriminant coefficients in the probit and logit discriminant models. Since, using the Wald chi-square statistic at the significance level of 5%, the significance probability of all the discriminant coefficients including the intercept are small, it is confirmed that the discriminant coefficients are very significant variables to distinguish the two groups in the probit and logit discriminant model. Table 2. The significance test of the model Analysis of Maximum Likelihood Estimates (Probit Model and Logit Discriminant Model) Variables Wald - χ 2 Pr > ChiSq Wald - χ 2 Pr > ChiSq Intercept CUR CR QR NR TA SA Secondary discriminant model The confusion matrix by the secondary discriminant function is shown in Table out of 89 companies belonging to healthy companies were correctly classified, and 63 out of 76 companies belonging to bad companies were correctly classified. Therefore, the misclassification rate for the total data is 23.03%, and the misclassification rate is 22.60% if the prior probability is equal. Quadratic Discriminant Function Table 3. Confusion matrix Bad Bad Rate Probit model and logit discriminant model Table 4 is the confusion matrix of the probit model and the logit discriminant model for evaluating the fit of the model. By the probit model, 71 out of 89 companies belonging to healthy companies were correctly classified, and 61 out of 76 companies belonging to bad companies were correctly classified. By the logit discriminant model, 71 out of 89 companies belonging to healthy companies were correctly classified, and 63 out of 76 companies belonging to bad companies were correctly classified. Therefore, the misclassification rates for the total data are 20% and 18.78% respectively, and the misclassification rates are 19.95% and 18.65% respectively when the classification reference value is 0.5. Probit Model Table 4. Confusion matrix Bad Bad Rate Logit Discriminant Model Bad Bad Rate

4 Association of Predicted Probabilities and Observed Responses Table 5 is statistics for evaluating the predictive power of the probit model and the logit discriminant model. The result of comparison between the two models shows that the values of the rank correlation coefficients of the logit discriminant model are higher overall than those of the probit model. Therefore, the estimated research model that predicts the financial risk is as follows. log ( p i ) = CUR CR 1 p i QR NR TA SA Table 5. Analysis of association of predicted probabilities and observed responses Probit Model Rank correlation coefficients Logit Discriminant Model Somer D Somer D Gamma Gamma Tau-a Tau-a C C This study derived a scientific model that predicts the financial risk of a firm. The results are summarized as follows. When the prior probability and the classification reference value are 0.5, the misclassification rate of the secondary discrimination model is 22.6%, the probit model 19.95%, and the logit discriminant model 18.65%. The logit discriminant model is the most suitable model to identify the financial risk, and its predictive power is also high as 88.8%. The results of this study are based on empirical data. Therefore, we expect that it will be used to establish the management strategy to secure financial of mediumsized enterprises as well as large corporations in the situation that asset quality of corporate loans is expected to be pressured due to sluggish sales following sluggish economic growth and market interest rate hike. And it will be meaningful for companies that are trapped in the financial risk due to US interest rate hikes to reduce debt and manage through financial risk management. However, the forecasting model through static analysis of financial ratios is insufficient to reflect the rapid change of business environment and has a limitation that can not grasp the strategic behavior of the company. Therefore, for the accuracy and usefulness of the forecasting model, dynamic analysis such as cash flow analysis, non-financial information analysis which can reflect the change of management environment, and studies of causal relationship model, AHP model (Analytical Hierarchy Process Model), data mining, and hybrid model etc. will be needed. ACKNOWLEDGMENTS This research was supported by research funding of Seokyeong University in REFERENCES [1] West, Robert C.,(1985), A factor-analytic approach to bank condition, Journal of Banking and Finance, Volp, pp [2] Tam, Kar Yan.(1991), Neural network models and the prediction of bank bankruptcy, Omega, Vol19, pp [3] Cho, M.,(1994), Predicting business failure in the hospitality industry : An application of logit model, Unpublished doctoral dissertation, Virginia Polytechnic Institute and State University. [4] Lee, Kun Chang, Ingoo Han, and Youngsig Kwon,(1996), Hybrid neural network models for bankruptcy predictions, Decision Support Systems, Vol18, pp [5] Davalos, S., Gritta, R. D. and Chow, G.(1999), The application of a neural network approach to predicting bankruptcy risks facing the major US air carriers, Journal of Air Transport Management, Vol5, pp [6] Altman, Edward I., R. Haldeman and P. Narayanan(1997), Zeta Analysis: A New Model to Identity Bankruptcy Risk of Corporation, Journal of Banking and Finance, pp [7] Davalos, S., R. Gritta & B. Adrangi,(2007), Deriving rules for forecasting air carrier financial stress and insolvency: A generic algorithm approach, Journal of the Transportation Research Forum, Vol46. [8] Boyacioglu, Mel 다 A., Yakup Kara., and Omer Kaan Baykan,(2009), Predicting bank financial failures using neural network, support vector machines and multivariate statistical methods: A comparative analysis in the sample of savings deposit insurance fund(sdif) transferred banks in Turkey, Expert Systems with Applications, Vol36, pp [9] Kim, S.,(2011), Prediction of hotel bankruptcy using support vector machine, artificial neural network, logistic regression, and multivariate discriminant analysis, The Service Industries Journal, Vol31, [10] Iturriaga, Felix J. Lopez, and Ivan P. Sanz,(2015), Bankruptcy visualization and prediction using neural 9922

5 networks: A study of US commercial banks, Expert Systems with Applications, Vol42, pp [11] An Chang-Ho,(2016), Study on privately held business performance forecast model estimates, The Society of Convergence Knowledge Transactions, Vol4, pp [12] Kim Jung-ae,(2017), The effect of self-identity on body shape management, International Journal of Advanced Culture Technology, Vol5, pp [13] Hwang Kyu-Sung and Kim Jeong-Lae,(2017), A Study on Hepatitis Infection Risk of Funeral director related to wearing PPE, Vol3, pp [14] Tabachnick, B. G., & Fidell, L. S. (2006), Using multivariate statistics (5th ed). New York, NY: HarperCollins College. 9923